2 /* { dg-require-effective-target sync_long_long_runtime } */
3 /* { dg-require-effective-target pthread_h } */
4 /* { dg-require-effective-target pthread } */
5 /* { dg-options "-pthread -std=gnu99" } */
6 /* { dg-additional-options "-march=pentium" { target { { i?86-*-* x86_64-*-* } && ia32 } } } */
8 /* test of long long atomic ops performed in parallel in 3 pthreads
9 david.gilbert@linaro.org */
14 /*#define DEBUGIT 1 */
19 #define DOABORT(x,...) {\
20 fprintf (stderr, x, __VA_ARGS__); fflush (stderr); abort ();\
25 #define DOABORT(x,...) abort ();
29 /* Passed to each thread to describe which bits it is going to work on. */
31 unsigned long long count; /* incremented each time the worker loops. */
32 unsigned int thread; /* ID */
33 unsigned int addlsb; /* 8 bit */
34 unsigned int logic1lsb; /* 5 bit */
35 unsigned int logic2lsb; /* 8 bit */
38 /* The shared word where all the atomic work is done. */
39 static volatile long long workspace;
41 /* A shared word to tell the workers to quit when non-0. */
42 static long long doquit;
44 extern void abort (void);
46 /* Note this test doesn't test the return values much. */
50 struct threadwork *tw = (struct threadwork*)data;
51 long long add1bit = 1ll << tw->addlsb;
52 long long logic1bit = 1ll << tw->logic1lsb;
53 long long logic2bit = 1ll << tw->logic2lsb;
55 /* Clear the bits we use. */
56 __sync_and_and_fetch (&workspace, ~(0xffll * add1bit));
57 __sync_fetch_and_and (&workspace, ~(0x1fll * logic1bit));
58 __sync_fetch_and_and (&workspace, ~(0xffll * logic2bit));
62 long long tmp1, tmp2, tmp3;
63 /* OK, lets try and do some stuff to the workspace - by the end
64 of the main loop our area should be the same as it is now - i.e. 0. */
66 /* Push the arithmetic section upto 128 - one of the threads will
67 case this to carry accross the 32bit boundary. */
68 for (tmp2 = 0; tmp2 < 64; tmp2++)
70 /* Add 2 using the two different adds. */
71 tmp1 = __sync_add_and_fetch (&workspace, add1bit);
72 tmp3 = __sync_fetch_and_add (&workspace, add1bit);
74 /* The value should be the intermediate add value in both cases. */
75 if ((tmp1 & (add1bit * 0xff)) != (tmp3 & (add1bit * 0xff)))
76 DOABORT ("Mismatch of add intermediates on thread %d "
77 "workspace=0x%llx tmp1=0x%llx "
78 "tmp2=0x%llx tmp3=0x%llx\n",
79 tw->thread, workspace, tmp1, tmp2, tmp3);
82 /* Set the logic bits. */
83 tmp2=__sync_or_and_fetch (&workspace,
84 0x1fll * logic1bit | 0xffll * logic2bit);
86 /* Check the logic bits are set and the arithmetic value is correct. */
87 if ((tmp2 & (0x1fll * logic1bit | 0xffll * logic2bit
89 != (0x1fll * logic1bit | 0xffll * logic2bit | 0x80ll * add1bit))
90 DOABORT ("Midloop check failed on thread %d "
91 "workspace=0x%llx tmp2=0x%llx "
92 "masktmp2=0x%llx expected=0x%llx\n",
93 tw->thread, workspace, tmp2,
94 tmp2 & (0x1fll * logic1bit | 0xffll * logic2bit |
96 (0x1fll * logic1bit | 0xffll * logic2bit | 0x80ll * add1bit));
98 /* Pull the arithmetic set back down to 0 - again this should cause a
99 carry across the 32bit boundary in one thread. */
101 for (tmp2 = 0; tmp2 < 64; tmp2++)
103 /* Subtract 2 using the two different subs. */
104 tmp1=__sync_sub_and_fetch (&workspace, add1bit);
105 tmp3=__sync_fetch_and_sub (&workspace, add1bit);
107 /* The value should be the intermediate sub value in both cases. */
108 if ((tmp1 & (add1bit * 0xff)) != (tmp3 & (add1bit * 0xff)))
109 DOABORT ("Mismatch of sub intermediates on thread %d "
110 "workspace=0x%llx tmp1=0x%llx "
111 "tmp2=0x%llx tmp3=0x%llx\n",
112 tw->thread, workspace, tmp1, tmp2, tmp3);
116 /* Clear the logic bits. */
117 __sync_fetch_and_xor (&workspace, 0x1fll * logic1bit);
118 tmp3=__sync_and_and_fetch (&workspace, ~(0xffll * logic2bit));
120 /* The logic bits and the arithmetic bits should be zero again. */
121 if (tmp3 & (0x1fll * logic1bit | 0xffll * logic2bit | 0xffll * add1bit))
122 DOABORT ("End of worker loop; bits none 0 on thread %d "
123 "workspace=0x%llx tmp3=0x%llx "
124 "mask=0x%llx maskedtmp3=0x%llx\n",
125 tw->thread, workspace, tmp3, (0x1fll * logic1bit |
126 0xffll * logic2bit | 0xffll * add1bit),
127 tmp3 & (0x1fll * logic1bit | 0xffll * logic2bit | 0xffll * add1bit));
129 __sync_add_and_fetch (&tw->count, 1);
131 while (!__sync_bool_compare_and_swap (&doquit, 1, 1));
139 /* We have 3 threads doing three sets of operations, an 8 bit
140 arithmetic field, a 5 bit logic field and an 8 bit logic
141 field (just to pack them all in).
145 |...,...|...,...|...,...|...,...|...,...|...,...|...,...|...,...
146 - T0 -- T1 -- T2 --T2 -- T0 -*- T2-- T1-- T1 -***- T0-
147 logic2 logic2 arith log2 arith log1 log1 arith log1
154 struct threadwork tw[3]={
155 { 0ll, 0, 27, 0, 56 },
156 { 0ll, 1, 8,16, 48 },
157 { 0ll, 2, 40,21, 35 }
160 pthread_t threads[3];
162 __sync_lock_release (&doquit);
164 /* Get the work space into a known value - All 1's. */
165 __sync_lock_release (&workspace); /* Now all 0. */
166 tmp = __sync_val_compare_and_swap (&workspace, 0, -1ll);
168 DOABORT ("Initial __sync_val_compare_and_swap wasn't 0 workspace=0x%llx "
169 "tmp=0x%llx\n", workspace,tmp);
171 for (t = 0; t < 3; t++)
173 err=pthread_create (&threads[t], NULL , worker, &tw[t]);
174 if (err) DOABORT ("pthread_create failed on thread %d with error %d\n",
181 __sync_lock_test_and_set (&doquit, 1ll);
183 for (t = 0; t < 3; t++)
185 err=pthread_join (threads[t], NULL);
187 DOABORT ("pthread_join failed on thread %d with error %d\n", t, err);
190 __sync_synchronize ();
192 /* OK, so all the workers have finished -
193 the workers should have zero'd their workspace, the unused areas
194 should still be 1. */
195 if (!__sync_bool_compare_and_swap (&workspace, 0x040000e0ll, 0))
196 DOABORT ("End of run workspace mismatch, got %llx\n", workspace);
198 /* All the workers should have done some work. */
199 for (t = 0; t < 3; t++)
201 if (tw[t].count == 0) DOABORT ("Worker %d gave 0 count\n", t);